Hi all,
this (quite long) patch addresses following issues in dwarf2cfi.c:
1) Read both .debug_frame and .eh_frame if they are present. Not all 
frames must be covered in both frames. Example: imagine an exec linked 
from file1.o that was compiled with -g and file2.o that was compiled 
without it. Than old gdb would parse only a .debug_frame that came from 
file1.o and wouldn't have any frame info for code from file2.o that 
provided only the .eh_frame. That's the reason why to parse both of them.
2) Of course we shouldn't have duplicate FDEs (one from .debug_frame and 
one from .eh_frame). If the file contains only an .eh_frame it's easy, 
there can't be duplicates, but if we parse both sections, we must check 
fde_chunks before adding new FDE from eh_frame. (see comments inside the 
patch).
3) The original dwarf2cfi.c didn't implement CIE Augmentation correctly. 
The main problem was a non-standard addressing (aug. 'R') that I had to 
rewrite completely. This fixes many problems with backtraces through 
relocatable (-fPIC compiled) objects without .debug_info.
4) parse_frame_info now also recognises aug. 'L' (= LSDA pointer present 
in FDE) but I was told I don't need LSDA in GDB, so I ignore its content.

2002-05-31  Michal Ludvig  <mludvig@suse.cz>

         * dwarf2cfi.c (dwarf_frame_buffer): Delete.
         (read_encoded_pointer): Add new argument,
         warn on indirect pointers.
         (execute_cfa_program): Warn on relative addressing,
         change messages.
         (frame_state_for, cfi_get_saved_register)
         (cfi_virtual_frame_pointer): Change message.
         (dwarf2_build_frame_info): Only call parse_frame_info
         for .debug_frame and .eh_frame.
         (parse_frame_info): New, derived from dwarf2_build_frame_info,
         fixed augmentation handling, added relative addressing,
         ignore duplicate FDEs. Added comments.

OK to commit?

Michal Ludvig
-- 
* SuSE CR, s.r.o     * mludvig@suse.cz
* +420 2 9654 5373   * http://www.suse.cz